The Science Behind Lie Detection Services

No scientific instrument directly detects a “lie”. Polygraphs measure physiological responses such as respiration, cardiovascular activity and electrodermal changes. Eye-tracking or ocular-motor systems measure patterns such as pupil response, fixation and reading behaviour. Software then applies a protocol or model to decide whether the pattern is more consistent with a truthful or deceptive condition. The quality of that conclusion depends on the test design, population, question format, data quality and validation evidence.

At Trojan Investigations, we believe lie-detection services must be explained without “truth machine” language. A structured test can be useful as one source of information, but it should not replace corroborating evidence, professional interviewing, document review or a proper investigation.

Originally published: 12th January 2024  |  Updated: 7th September 2026

Why “Lie Detector” Is a Misleading Name

The popular phrase suggests a machine can look inside a person’s mind and identify falsehood with certainty. That is not how any established method works. Human deception is not a single physiological event. People can be anxious while truthful, calm while deceptive, confused by a question, worried about consequences or affected by medication, fatigue and individual differences.

Scientific evaluation therefore focuses on the performance of a specific protocol under specified conditions: sensitivity, specificity, false positives, false negatives, inconclusive results and whether laboratory findings generalise to real-world settings.

How a Polygraph Works

A polygraph records physiological channels while an examiner asks a structured set of questions. Different protocols compare responses to relevant, comparison or concealed-information stimuli in different ways. The result is based on changes in measured physiology and a scoring system, not on a sensor that recognises dishonesty directly.

The long-running scientific debate is therefore about validity: how reliably those physiological patterns distinguish deceptive from truthful people across different settings and populations.

What the National Academies Review Found

The US National Research Council’s major consensus review, The Polygraph and Lie Detection, remains an influential scientific assessment. It concluded that polygraph testing has some ability to discriminate deception from truth in specific-event testing, but that accuracy is well below perfection and the evidence base is affected by methodological limitations. It was particularly critical of using polygraph screening to detect rare security threats across large populations.

The lesson for private clients is important: an accuracy percentage should never be detached from the protocol, population and decision being made.

How Polygraphs Are Used in England and Wales

The UK government uses mandatory polygraph examinations in a defined probation and offender-management context for certain people on licence. The Ministry of Justice policy framework was updated in January 2026, and the Crime and Policing Act 2026 expanded the framework for specified offender groups.

This official use is often misunderstood. GOV.UK explains that polygraph is an additional source of risk-management information. Results are not simply treated as courtroom proof of guilt. Government guidance states that the result cannot be used in criminal proceedings against the released person and that a significant response is not, by itself, the same as proving a new offence.

What Is Ocular-Motor or Eye-Tracking Lie Detection?

Eye-tracking approaches record involuntary or semi-involuntary ocular behaviour while a person completes a computer-administered task. Depending on the protocol, the system may analyse pupil dilation, fixation duration, reading time, saccades or patterns associated with recognising concealed information.

The underlying theory is that deception, concealment or recognition can alter cognitive workload and attention, which can in turn affect eye behaviour. This is a different measurement strategy from a polygraph, but it still produces an inference rather than reading a lie directly.

What Does the Newer Eye-Tracking Research Show?

Research is active and results vary by paradigm. A 2025 peer-reviewed mock-crime study of an ocular-motor multiple-issue screening protocol reported group classification rates ranging from 83.3% to 91.7% and item-level classification between 81.1% and 88.3% across the tested topics.

A 2026 Scientific Reports study compared an experimental eye-tracking Concealed Information Test with a polygraph under the same controlled conditions. The eye-tracking model achieved 94.9% accuracy and the polygraph 88.2% among the analysed participants, while some polygraph recordings were inconclusive.

Those are interesting findings, but they do not mean every eye-tracking test has 94.9% real-world accuracy. The study used a specific concealed-information paradigm, participant sample, controlled design and machine-learning classifier. Generalising beyond that requires independent replication and careful validation.

What About EyeDetect and Similar Commercial Systems?

Commercial ocular-motor systems such as EyeDetect use computer-administered testing and algorithms to classify response patterns. The vendor publishes peer-reviewed and other research supporting its protocols. Independent peer-reviewed research also shows that eye behaviour can carry information relevant to deception or concealed knowledge under some tasks.

However, a responsible provider should distinguish vendor claims from independent evidence and explain what a particular test is validated to assess. We do not describe any system as capable of reading minds, proving guilt or replacing evidence.

Why Eye Movements Cannot Be Read Like Body-Language Myths

There is an important difference between instrumented eye tracking and popular claims such as “looking up to the right means someone is lying”. A peer-reviewed 2012 study found no support for the Neuro-Linguistic Programming claim that particular gaze directions reliably indicate lying.

Scientific eye-tracking protocols rely on measured features and defined tasks, not a simple visual rule that an interviewer can apply by watching someone’s eyes.

What About Voice Stress Analysis?

Voice-stress systems attempt to identify acoustic changes associated with stress or deception. The scientific challenge is the same one faced by many credibility tools: stress is not synonymous with lying. Voice can change because of fear, illness, language, fatigue, emotion, microphone quality and many other factors.

Any provider presenting voice stress as certain proof of dishonesty should be treated cautiously.

Can Artificial Intelligence Detect Lies?

AI can classify patterns in text, voice, facial movement, physiology and eye behaviour, but the model is only as good as the underlying theory, labels and validation. A 2024 review in Trends in Cognitive Sciences warned that many AI deception-detection approaches have focused on technical performance without a sufficiently strong methodological and theoretical foundation.

A separate 2024 peer-reviewed critique of machine-learning polygraph approaches argued that weak labels and unresolved validity problems can make apparently sophisticated models scientifically misleading.

Why Nervousness Matters

A truthful person can be highly anxious because they fear being disbelieved, losing a job or relationship, or facing a serious allegation. Conversely, a deceptive person may be calm. This is why responsible interpretation does not equate “strong physiological response” with “lie” without reference to the protocol.

Question wording, comprehension and the person’s ability to participate meaningfully also matter.

What Makes a Credibility Test More Defensible?

Can Lie Detection Replace an Investigation?

No. In a relationship, workplace or commercial dispute, a test result should be considered alongside documents, timelines, witness evidence, digital records and other lawful facts. Our personal investigations and business investigations focus on corroboration rather than treating a single instrument as the whole case.

If the matter is connected to litigation, the instructing solicitor should decide what evidential value any test may have. Our legal and litigation support can focus on evidence that can be independently verified.

When Can a Lie-Detection Test Be Useful?

A structured test may help narrow an investigation, encourage disclosure, identify questions requiring corroboration or support a voluntary internal fact-finding process. It can also be useful where all parties understand the limitations and agree in advance how the result will be interpreted.

It is much less defensible when used as the sole basis for a high-stakes accusation, employment dismissal, criminal allegation, family decision or public claim about a person’s honesty.

Consent, Fairness and Data Protection

A credibility test processes sensitive behavioural or physiological information and can have serious consequences for the participant. Participation should be genuinely informed, the purpose should be clear, questions should be relevant and data should be retained only as long as necessary.

In employment settings, power imbalance is especially important. “Consent” may not always be the most appropriate data-protection basis simply because an employee signed a form. Organisations need to consider lawful basis, fairness, necessity and proportionality.

How We Would Explain a Result

False Positives, False Negatives and Base Rates Matter

Any credibility test can make two broad types of error. A false positive classifies a truthful person as deceptive; a false negative classifies a deceptive person as truthful. The practical impact of those errors depends on the decision being made. In a relationship dispute, a false positive can destroy trust. In employment, it can unfairly damage a person’s career. In security screening, even a test with apparently strong accuracy can generate many false alarms when the behaviour being searched for is rare.

This is the “base-rate” problem highlighted in scientific reviews of screening. If only a very small proportion of a large tested population is actually deceptive about the target issue, a method that is less than perfect may still identify more innocent people than true offenders. That is one reason controlled specific-issue testing and broad population screening should not be treated as scientifically equivalent.

Independent Replication Is More Valuable Than a Marketing Percentage

A strong evidence base is built when different research groups can test the same underlying method, publish methods and limitations, and obtain broadly consistent results. Vendor-funded studies can still be scientifically useful, but clients should know whether a headline accuracy claim comes from the manufacturer, an independent research team, a laboratory mock-crime design or real-world field testing.

For us, the responsible position is to present the test within those limits and to avoid converting a promising research result into a guarantee about one individual case.